The work aims to propose an image encryption algorithm based on improved generalized Arnold mapping to improve the security of the encrypted image regarding the problem that the image scrambling effect of Arnold mapping is poor and the chosen plaintext attack cannot be resisted. The algorithm introduced Kent mapping to generate control parameters. First, the images were partitioned to reduce the image correlation. Each image’s Arnold mapping position was scrambled. Then the 2D logistic mapping was used to generate chaotic sequences. Through the hyperbolic tangent transformation processing, XOR was conducted according to the parity sequence and the scrambled images, and the final ciphertext image was obtained. The information entropy of the ciphertext image obtained through the algorithm was 7.9899, the pixel change rate was 99.61%, and the uniform average change degree was 30.73%. The algorithm is simple and easy to implement. It can effectively resist all kinds of statistical attacks and plaintext attacks. It is also characterized by large key space, good scrambling effect and strong robustness, etc.